remote_worker.py

   1 #!/usr/bin/env python3
   2
   3 """A simple utility to unpickle some code, run it, and pickle the
   4 results.
   5 """
   6
   7 import logging
   8 import os
   9 import signal
  10 import threading
  11 import sys
  12 import time
  13
  14 import cloudpickle  # type: ignore
  15 import psutil  # type: ignore
  16
  17 import argparse_utils
  18 import bootstrap
  19 import config
  20 from thread_utils import background_thread
  21
  22
  23 logger = logging.getLogger(__file__)
  24
  25 cfg = config.add_commandline_args(
  26     f"Remote Worker ({__file__})",
  27     "Helper to run pickled code remotely and return results",
  28 )
  29 cfg.add_argument(
  30     '--code_file',
  31     type=str,
  32     required=True,
  33     metavar='FILENAME',
  34     help='The location of the bundle of code to execute.',
  35 )
  36 cfg.add_argument(
  37     '--result_file',
  38     type=str,
  39     required=True,
  40     metavar='FILENAME',
  41     help='The location where we should write the computation results.',
  42 )
  43 cfg.add_argument(
  44     '--watch_for_cancel',
  45     action=argparse_utils.ActionNoYes,
  46     default=True,
  47     help='Should we watch for the cancellation of our parent ssh process?',
  48 )
  49
  50
  51 @background_thread
  52 def watch_for_cancel(terminate_event: threading.Event) -> None:
  53     logger.debug('Starting up background thread...')
  54     p = psutil.Process(os.getpid())
  55     while True:
  56         saw_sshd = False
  57         ancestors = p.parents()
  58         for ancestor in ancestors:
  59             name = ancestor.name()
  60             pid = ancestor.pid
  61             logger.debug(f'Ancestor process {name} (pid={pid})')
  62             if 'ssh' in name.lower():
  63                 saw_sshd = True
  64                 break
  65         if not saw_sshd:
  66             logger.error(
  67                 'Did not see sshd in our ancestors list?!  Committing suicide.'
  68             )
  69             os.system('pstree')
  70             os.kill(os.getpid(), signal.SIGTERM)
  71             time.sleep(5.0)
  72             os.kill(os.getpid(), signal.SIGKILL)
  73             sys.exit(-1)
  74         if terminate_event.is_set():
  75             return
  76         time.sleep(1.0)
  77
  78
  79 @bootstrap.initialize
  80 def main() -> None:
  81     in_file = config.config['code_file']
  82     out_file = config.config['result_file']
  83
  84     stop_thread = None
  85     if config.config['watch_for_cancel']:
  86         (thread, stop_thread) = watch_for_cancel()
  87
  88     logger.debug(f'Reading {in_file}.')
  89     try:
  90         with open(in_file, 'rb') as rb:
  91             serialized = rb.read()
  92     except Exception as e:
  93         logger.exception(e)
  94         logger.critical(f'Problem reading {in_file}.  Aborting.')
  95         stop_thread.set()
  96         sys.exit(-1)
  97
  98     logger.debug(f'Deserializing {in_file}.')
  99     try:
 100         fun, args, kwargs = cloudpickle.loads(serialized)
 101     except Exception as e:
 102         logger.exception(e)
 103         logger.critical(f'Problem deserializing {in_file}.  Aborting.')
 104         stop_thread.set()
 105         sys.exit(-1)
 106
 107     logger.debug('Invoking user code...')
 108     start = time.time()
 109     ret = fun(*args, **kwargs)
 110     end = time.time()
 111     logger.debug(f'User code took {end - start:.1f}s')
 112
 113     logger.debug('Serializing results')
 114     try:
 115         serialized = cloudpickle.dumps(ret)
 116     except Exception as e:
 117         logger.exception(e)
 118         logger.critical(f'Could not serialize result ({type(ret)}).  Aborting.')
 119         stop_thread.set()
 120         sys.exit(-1)
 121
 122     logger.debug(f'Writing {out_file}.')
 123     try:
 124         with open(out_file, 'wb') as wb:
 125             wb.write(serialized)
 126     except Exception as e:
 127         logger.exception(e)
 128         logger.critical(f'Error writing {out_file}.  Aborting.')
 129         stop_thread.set()
 130         sys.exit(-1)
 131
 132     if stop_thread is not None:
 133         stop_thread.set()
 134         thread.join()
 135
 136
 137 if __name__ == '__main__':
 138     main()